Natural voice translation with Gemini 3.5 Live Translate

Twenty years ago, translation at Google began as one of our pioneering machine learning experiments. This week we took our next big leap with Gemini 3.5 Live Translate. This new audio model for live speech-to-speech translation automatically detects more than 70 languages while preserving the speaker's natural intonation and pacing, eliminating awkward pauses. You can now have fluid, near-real-time conversations during multilingual calls, meetings or travel, removing language barriers in seconds. It's rolling out across Google Translate, Gemini Live API and Google AI Studio.

Gemma is our family of open AI models, and Gemma 4 is our most capable version yet — allowing developers to run it on their own hardware to build great apps and experiences. In fact, the Gemma 4 model family has officially crossed 150 million downloads. With so many people building, we wanted to highlight a few projects that caught our eye. See how three builders are using Gemma 4 to push creative boundaries and create entirely new experiences.